Environment-adaptive machine learning potentials

Machine learning interatomic potentials (MLIP) have shown great promise in predicting material properties when tailored for specific applications in mind. However, they face limitations when expanded to capture a diverse set of local atomic environments. Here, the authors introduce a novel method to construct environment-adaptive machine-learned (EAML) interatomic potentials by adapting to the local atomic environment of each atom. The approach shows a significant improvement in the accuracy of the EAML potential compared to the current state-of-the-art, paving the way for a new class of MLIPs.